dane-pgp

The article mentions that CSP would be a complete defence against this attack, except that there's a "prefetch trick" that bypasses that defence in Chrome.

Coincidentally, last month the Chrome team decided to abandon[0] plans to implement the prefetch-src CSP directive, in favour of a "least-restrictive" directive[1].

The new approach is being tracked on Chrome's feature status website[2], which says: "This allows developer to use resource hints without needing to tweak their content security policy, while giving a tool to prevent exfiltration by having default-src block prefetches."

[0] https://bugs.chromium.org/p/chromium/issues/detail?id=801561

[1] https://bugs.chromium.org/p/chromium/issues/detail?id=140644...

[2] https://chromestatus.com/feature/5553640629075968

> who’s to say that what you see in /lib/package.min.js is the real result of minifying /src/package.js?

This is why NPM really needs to start demanding that packages (with more than N downloads per month) are reproducibly buildable from source.

Someone did actually produce a tool for performing these checks[0], specifically with the linked article in mind, but unfortunately the online service connected to the tool has been discontinued.

[0] https://medium.com/hackernoon/what-if-we-could-verify-npm-pa...

I don't know why you are being silently downvoted, as I think it is worth talking about the potential of using static analysis to improve things.

One promising approach is Endo[0] which "uses LavaMoat to automatically generate reviewable policies that determine what capabilities will be distributed to third party dependencies."

[0] https://github.com/endojs/endo

Or write an AI that does sentiment analysis on the Twitch chat, to find those moments automatically.

So maybe people with a net worth above 2.2 million should be paying a wealth tax instead of an income tax.

A better comparison might be:

I am not going to build a chainsaw, regardless of whether someone else could commit a stabbing.

You may want to read something like "Common misconceptions about Germany's energy transition"[0] for a fuller picture of the situation there. In particular, it points out that between 2002 and 2021:

"both hard and lignite [coal usage] decreased from 251.9 TWh to 146.5 TWh for a reduction of 109.4 TWh. So Germany reduced both coal and nuclear at the same time, coal did not grow, and the coal reduction was actually larger than the nuclear reduction."

[0] https://chadvesting.substack.com/p/common-misconceptions-abo...

The task force's original top priority was convincing the world that they don't exist, but they apparently gave up any pretence of that last year when they rewarded Lennart Poettering with a job at the company.

Because we don't have a global minimum wealth tax.

Countries have shown that they can work together to implement a global minimum corporation tax[0] to prevent the race-to-the-bottom / tragedy-of-the-commons dynamics of corporations playing countries off against each other, so why can't countries now do the same for taxes on the wealth of ultra-millionaires?

I would base it on Elizabeth Warren's plan[1] for a national wealth tax in the US, which would only apply to the top 0.1% of households there.

[0] https://en.wikipedia.org/wiki/Global_minimum_corporate_tax_r...

[1] https://elizabethwarren.com/plans/ultra-millionaire-tax

If you think that's bad, you're not going to like the fact that they're letting the NSA design a system for "TPM-based Network Device Remote Integrity Verification":

https://www.ietf.org/archive/id/draft-ietf-rats-tpm-based-ne...

> a trusted party reviewing code & posting content integrity hashes

That sounds like some combination of Crev[0], Sigstore[1], and Trillian[2].

[0] https://github.com/crev-dev/cargo-crev/blob/master/cargo-cre...

[1] https://docs.sigstore.dev/

[2] https://transparency.dev/

This is why you have to protect your Hardware Security Module with a passphrase that is kept only in your brain. Compelling the disclosure of that passphrase is much more likely to violate the Fifth Amendment, especially if you set your passphrase to something like "I killed him and buried the body under my garage".

Of course such an important passphrase shouldn't only exist in the head of one person, and instead should be distributed between multiple members of the company (so perhaps the HSM could require N of M passphrases to unlock the master secret, using Shamir's Secret Sharing), which means creating a very complicated on-going criminal conspiracy, with new hires forced to further the crime in unique and creative ways, so that their individually-chosen passphrases can't be guessed.

> The alternative is just to leave the warrant canary and live happily after.

Another alternative would be to implement Binary Transparency, and make the app only download updates whose hashes appear in an independently-run jurisdictionally-decentralized append-only log. (Rolling out such a change might take too long to help the target of the current NSL, but it would protect future users, and announcing such a feature would itself be sending quite an important message).

I suppose if your business relies on keeping its source code secret, then you could just put an "if userName == the_target_mentioned_in_the_NSL" branch into the code, so that all your users receive the same update, but hopefully someone out there would be able to reverse engineer that code (perhaps after an anonymous tip-off).

Perhaps the government would be willing to pay for a software engineer to obfuscate the code enough that this malicious branch won't be detected in time, but I think that would put selective pressure on software companies to not distribute obfuscated binaries.

What would "tested in the US courts" look like? Would the federal government take an entity to court, demanding that they put their warrant canary back up, as a form of compelled speech?

No, I don't think there's been an example of that happening, but perhaps we would expect that if the government wanted to bring such a case, they would use a secret court, or an NSL, which would likely result in the warrant canary being added back to the site before anyone noticed. (An injunction against removing the canary could even be included in the initial secret warrant).

Don't hate the player, hate the game?

When are these jurisdictions going to start demanding that Android phones not allow side-loading? And when will Western countries start doing the same?

Fortunately the EU seems to be pushing back on governments' desires to restrict general purpose computing, but I'm not sure how long that will last.

I wonder if Dmitry Medvedev is busy right now?

> Mass murder and subjugation at scale, which were perpetrated by many governments in the 20th century

I've got bad news for you, buddy...

> Neither of those is possible, or will be possible, in the DNS PKI.

Transparency logs for domain issuance is completely possible, it just requires some engineering and deployment. Remember that HTTPS was in use for decades before CT logging of certificates became mandatory.

More importantly, though, we need to be clear about what "noncompliance" means. In the web PKI, it means a CA issuing a certificate for a domain that the requester doesn't control, but the equivalent for that in the DNS PKI would be TLD .foo publishing data for a domain example.bar which is not an attack at all, because no one would care what .foo thinks about a .bar domain.

So, rather than relying on browsers being brave enough to kill a large provider (and we need to be honest with ourselves about how much leeway browsers would give to Let's Encrypt if they ever suffered a catastrophic security breach), the DNS PKI simply isn't vulnerable to this problem, because TLDs can't issue certs for domains that aren't registered under them.

DNSSEC doesn't protect you against the American government if you have a .org domain, but I doubt an American court could give Microsoft control over a domain registered under a ccTLD like .de or .ru or .za for example.

I suspect Microsoft would also have trouble taking control of a domain registered under a gTLD run by a company based outside the US, but it would be interesting to see how the agreements between the gTLDs and ICANN would work out in practice.

> they really don't have overall goals or motivations ... Their output is just an average of what text would some sequence of text

Yes, but "Optimality is the tiger, and agents are its teeth".[0]

I don't want to spoil that essay by explaining how the LLM suddenly starts acting like an agent, but I can assure you that the author does a very good job of setting up a "Yes, that seems safe to me" thought experiment before revealing the "Oh no, that's terrible!" outcome.

[0] https://www.alignmentforum.org/posts/kpPnReyBC54KESiSn/optim...

Isn't it obvious? The person playing the AI just has to ask the Gatekeeper: "Would you lie to save a life?" and "Do you think that a fear-generating outcome to this thought experiment will have more than a one-in-a-million chance of increasing the chance of the AI-not-kill-everyone scenario by 0.1 percent?".

The sorts of rationalists who would play the Gatekeeper would probably answer yes to both of those questions, and draw the obvious conclusion that "losing" in their role would have an expected outcome of at least one life saved. If they don't value Truth (with a capital T) then there is no reason not to pretend that there really is some amazing "AI convinced me to let it out of the box" secret argument.

The alternative for Apple is some sort of "malicious compliance", where they offer "Insecure Messages" as a UK-alternative to their Messages app.

It should prominently display an uncloseable banner at the top of the app saying "Due to a surveillance law passed by the Tory government, Apple is not allowed to secure any messages you send or receive in this app".

Users in other countries who communicate with UK users should have a similar but minimisable message, warning them of the same problem.

> a follow-up IETF project to use MLS for inter-messenger interoperability

It's worth highlighting (as a comment in the linked discussion does) that the EU's newly-passed Digital Markets Act requires interoperability between the messaging apps of large "gatekeeper" platforms.

If Twitter does end up supporting E2EE messaging, it may soon be forced to implement MLS, and if Facebook and Google are really supportive of MLS then soon(TM) most people online will be able to communicate via that technology.

> Under the Online Safety Bill, which is currently being debated in the House of Commons, companies will be compelled to weaken security and provide “backdoor access” that bypasses encryption and provide access to any encrypted data in messages, cloud storage or logs on request.

If companies are required to provide backdoors, presumably what this means in practice is that they release a targeted update to their app with extra code in which breaks the privacy guarantees just for a specific user.

Assuming the app is open source and reproducibly builds, app stores could protect users by implementing Binary Transparency with a public append-only log, but presumably the law will be flexible enough to apply to OS vendors too, forcing them lie about the hashes of the apps that are downloaded.

> multiplying two 5-digit numbers

To be fair, 99% of humans can't do that unaided (that is, without a calculator, or some way of visually laying out the intermediate steps) either. In fact, I'd say that 90% of humans can't do the necessary long multiplication steps even with a pencil and paper, and maybe 20% of humans couldn't do it even with a calculator (due to being too young; or having a disability; or growing up in a environment where they are not taught how to use a calculator or even the concept of multiplication at all).

I just think it's ironic if people are suggesting that their threshold for finding an AI "impressive" requires it to already be effectively super-human. It's like they're retreating from the bailey of long AI timelines to the motte of: "Sure, I can't think of anything a human can do that an AI can't, but that's not really significant, because humans aren't particularly intelligent".

Perhaps a more relevant test would be to ask ChatGPT something like: "If a human wanted to calculate the product of the two numbers 12345 and 56789, and they had a calculator with them, what sequence of buttons would they press, in order, to find the result?".

> Say "I don't know."

That's a good answer, thank you. A similar answer might be: "Admit it's wrong". Technology really has gone in a strange direction if we end up measuring the performance of software systems in terms of which personality flaws they have.

I think this was actually a really thought-provoking question, when it was asked a few months ago, but now I'm wondering whether the guesses given have already been proven wrong by ChatGPT, or at least whether people would give guesses today that are very different from those given pre-ChatGPT.

I think what many societies have discovered in the past few years is, if a proposed authoritarian policy only (or just disproportionately) hurts "the other side", then half the country will abandon any principles they used to have so they can claim that it's actually a good policy to implement.

I suppose someone could have the philosophical position that women have a moral duty to turn all their egg cells into human lives, but just because we can imagine such a biologically impossible position doesn't mean that you can declare all positions other than your own to be completely unreasonable.

There is a much brighter dividing line between gametes and embryos than between pre-viability and post-viability embryos, so just on the basis of legal certainty it would be better to say that life starts at conception. I also think that it is philosophically parsimonious to define human beings as unified entities with distinct DNA from their parents, otherwise each skin cell should have its own human rights (which would mean that obese people get to cast more votes than skinny people).

In any case, given that there is ambiguity on where to draw the line, the safest approach is to pick the earliest practical point, since mistakenly protecting non-humans is better than mistakenly manslaughtering entities that we later concede were humans. If we take the view that every gamete has to become a human life, then nearly everyone in the country should be arrested as soon as they reach puberty.